DeepSeek sẽ huấn luyện mô hình V5 trên chip Huawei Ascend

The Huoxing Caijing report, citing a November 3 article from the Science and Technology Daily, states that according to multiple industry insiders, the AI company DeepSeek plans to train its next-generation large model, DeepSeek V5, using Huawei Ascend chips. Previously, Ascend chips were primarily used for inference tasks in DeepSeek’s large models. The report notes that at the end of September this year, DeepSeek open-sourced infrastructure components for the Huawei Ascend computing platform, including the TileLang high-level language compiler, computational libraries, and distributed communication libraries, each corresponding one-to-one with the previously open-sourced components for the NVIDIA platform. This report was written by journalist Huang Xinyi of the Science and Technology Daily.